Workday Case: AI doesn't invent biases. It learns them.

What happens when decisions that for years we made on autopilot, in silence, and often without registering them, begin to make themselves... and at a speed we can no longer keep up with?

The problem arises when AI learns from decisions made by people. Because if organizations have chosen talent for years based on biases—conscious or unconscious—artificial intelligence can amplify those exact same biases at a speed and scale we have never seen before. AI doesn't invent biases. It learns them.

And that is the sequence of events, hard as it is for us to admit: first comes the person, with their fears, lack of knowledge, biases, and prejudices. Making decisions. Generating information. Training artificial intelligence systems, almost always without knowing it. Afterward, those systems, fueled by that enormous living database, start deciding faster and faster... and at a scale we have never seen before.

Where our criteria come from

The criteria we use to make decisions carry a high degree of subjectivity. They bring along our life experience, what we have seen, what we believe, what we value, and what we don't. If we believe that young people with no work experience cannot take on certain roles, and that people over 50 cannot learn or adapt to certain demands, then our decisions—all of them, one by one—will reflect that belief precisely.

When that belief spreads, multiplies, and ends up being accepted as truth, a culture emerges. A culture that consolidates those values and ways of doing things, often without anyone questioning them. They replicate automatically and ensure their continuity over time. Until now, that "autopilot" had a limit: human capacity. In the case we are analyzing, a person can only read, evaluate, and discard a certain number of CVs per day.

What changes with technology

What changes is precisely that limit. Those decisions, repeated for years—often unconsciously—begin to turn into rules written as algorithms. They stop being applied case by case. They start being applied across the board. And at a speed that vastly exceeds human capacity to review, question, or correct each decision.

That is where—along with all the real advantages that speed, standardization, and the replacement of routine tasks bring (think of reading CVs and putting together an initial pool of candidates to interview)—a snowball effect begins to form. We can only see it if we recover, train, and exercise our critical thinking in the face of new technologies. If we question our own certainties. Even that comforting certainty that technology is going to replace us or solve all the work we don't want to do.

A case, with a name and numbers

Annie Coleman, global ambassador for the Stanford Center on Longevity, often uses a case that sums up this challenge. Workday, one of the most widely used Human Resources platforms in the world, is facing a class-action lawsuit for alleged age discrimination in its automated screening processes. According to the lawsuit, the platform rejected more than 1.1 billion applications during the analyzed period. The U.S. courts decided that the lawsuit should proceed as a class action, allowing anyone over 40 who applied through Workday since 2020 to join the claim if they believe they were affected by those automated processes.

For years, age discrimination in hiring processes was hard to prove. An interview, an undocumented decision, a "not a fit for the profile" that no one ever quite explained was enough. But when the decision shifts from a person to a platform, bias doesn't disappear: it scales up, leaving a trace and a trail. And "we didn't know" ceases to be a valid answer.

This is neither an isolated case nor a distant anecdote. It is evidence that a bias that remained implicit in an individual decision for years can now be codified and applied systematically, at a speed and magnitude that no recruitment department could ever achieve working "by hand".

Questioning ourselves, not confronting each other

The responsibility ahead of us does not lie solely with those who design algorithms. It belongs to all of us. We need to learn to work intergenerationally, interdisciplinarily, and interculturally. To complement one another. But above all, to question ourselves. Why do certain decisions seem obvious to us? Why do we rule out certain options almost automatically? What prejudices are we taking for granted without even realizing it?

That conversation, when honest, is worth more than any bias audit commissioned from a vendor. Because an audit can check the code. But the uncomfortable question forces us to check ourselves. And that is where, almost always, bias begins.

Learn, learn, learn. With direction, intention, and our feet on the ground.

And learning, in these times, is much more than downloading ChatGPT, asking a question, and marveling at the answer. It is understanding what lies behind what we see. And that requires us to be clearer-headed. Also slower. Because when running, it is very hard to see the details that appear when we walk.

Understand in order to later integrate. Broaden our perspective, open new dialogues, and listen to voices we disagree with or even don't understand yet. And from there, dare to transform. Seize the opportunities these new technologies bring—working better, living with greater health and well-being, knowing ourselves more deeply—while keeping our eyes open to the shadows they also entail.

Because the stakes are higher and the moment is now.

Artificial intelligence is already transforming how we as individuals are evaluated, chosen, or discarded. It is happening much more than we imagine. And precisely for that reason, it is up to us to examine our own biases. Because what technology amplifies will depend, to a large extent, on what we first decide to change in ourselves.